This paper presents findings from a large-scale study of Internet packet dynamics, focusing on 20,000 TCP bulk transfers between 35 Internet sites. The study uses a measurement framework where sites run special measurement daemons ("NPDs") to facilitate data collection. The analysis is based on measurements of TCP bulk transfers, which are more realistic and allow for fine-scale probing compared to fixed-rate UDP or ICMP packets. The paper characterizes unusual network behaviors such as out-of-order delivery, packet replication, and packet corruption. It discusses a robust receiver-based algorithm for estimating "bottleneck bandwidth," which is crucial for subsequent analysis of packet loss and delay. The study finds that out-of-order delivery is prevalent, with rates varying significantly between sites and often coinciding with route flutter. Packet replication and corruption are also observed, though less frequently. The paper introduces a more robust procedure, "packet bunch modes" (PBM), for estimating bottleneck bandwidth, which addresses the limitations of previous techniques like "packet pair." PBM can handle out-of-order delivery, clock resolution issues, and changes in bottleneck bandwidth, providing more accurate estimates. The study finds that PBM detects a single bottleneck 95-98% of the time and infers multi-channel bottlenecks in 1-2% of connections. The paper also examines packet loss patterns, finding that loss rates nearly doubled from 1994 to 1995, with higher loss rates observed during busy periods. The analysis distinguishes between data packet loss and acknowledgment (ack) packet loss, noting that ack losses provide a clearer picture of overall Internet loss patterns. The study concludes with a discussion of the implications of these findings for TCP performance and network design.This paper presents findings from a large-scale study of Internet packet dynamics, focusing on 20,000 TCP bulk transfers between 35 Internet sites. The study uses a measurement framework where sites run special measurement daemons ("NPDs") to facilitate data collection. The analysis is based on measurements of TCP bulk transfers, which are more realistic and allow for fine-scale probing compared to fixed-rate UDP or ICMP packets. The paper characterizes unusual network behaviors such as out-of-order delivery, packet replication, and packet corruption. It discusses a robust receiver-based algorithm for estimating "bottleneck bandwidth," which is crucial for subsequent analysis of packet loss and delay. The study finds that out-of-order delivery is prevalent, with rates varying significantly between sites and often coinciding with route flutter. Packet replication and corruption are also observed, though less frequently. The paper introduces a more robust procedure, "packet bunch modes" (PBM), for estimating bottleneck bandwidth, which addresses the limitations of previous techniques like "packet pair." PBM can handle out-of-order delivery, clock resolution issues, and changes in bottleneck bandwidth, providing more accurate estimates. The study finds that PBM detects a single bottleneck 95-98% of the time and infers multi-channel bottlenecks in 1-2% of connections. The paper also examines packet loss patterns, finding that loss rates nearly doubled from 1994 to 1995, with higher loss rates observed during busy periods. The analysis distinguishes between data packet loss and acknowledgment (ack) packet loss, noting that ack losses provide a clearer picture of overall Internet loss patterns. The study concludes with a discussion of the implications of these findings for TCP performance and network design.